Accurate Ab Initio Molecular Dynamics Research Articles

Perspective: Atomistic simulations of water and aqueous systems with machine learning potentials.

As the most important solvent, water has been at the center of interest since the advent of computer simulations. While early molecular dynamics and Monte Carlo simulations had to make use of simple model potentials to describe the atomic interactions, accurate abinitio molecular dynamics simulations relying on the first-principles calculation of the energies and forces have opened the way to predictive simulations of aqueous systems. Still, these simulations are very demanding, which prevents the study of complex systems and their properties. Modern machine learning potentials (MLPs) have now reached a mature state, allowing us to overcome these limitations by combining the high accuracy of electronic structure calculations with the efficiency of empirical force fields. In this Perspective, we give a concise overview about the progress made in the simulation of water and aqueous systems employing MLPs, starting from early work on free molecules and clusters via bulk liquid water to electrolyte solutions and solid-liquid interfaces.

CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling

Large-scale simulations with complex electron interactions remain one of the greatest challenges for atomistic modelling. Although classical force fields often fail to describe the coupling between electronic states and ionic rearrangements, the more accurate ab initio molecular dynamics suffers from computational complexity that prevents long-time and large-scale simulations, which are essential to study technologically relevant phenomena. Here we present the Crystal Hamiltonian Graph Neural Network (CHGNet), a graph neural network-based machine-learning interatomic potential (MLIP) that models the universal potential energy surface. CHGNet is pretrained on the energies, forces, stresses and magnetic moments from the Materials Project Trajectory Dataset, which consists of over 10 years of density functional theory calculations of more than 1.5 million inorganic structures. The explicit inclusion of magnetic moments enables CHGNet to learn and accurately represent the orbital occupancy of electrons, enhancing its capability to describe both atomic and electronic degrees of freedom. We demonstrate several applications of CHGNet in solid-state materials, including charge-informed molecular dynamics in LixMnO2, the finite temperature phase diagram for LixFePO4 and Li diffusion in garnet conductors. We highlight the significance of charge information for capturing appropriate chemistry and provide insights into ionic systems with additional electronic degrees of freedom that cannot be observed by previous MLIPs.

Machine learning force fields for molecular liquids: Ethylene Carbonate/Ethyl Methyl Carbonate binary solvent

Highly accurate ab initio molecular dynamics (MD) methods are the gold standard for studying molecular mechanisms in the condensed phase, however, they are too expensive to capture many key properties that converge slowly with respect to simulation length and time scales. Machine learning (ML) approaches which reach the accuracy of ab initio simulation, and which are, at the same time, sufficiently affordable hold the key to bridging this gap. In this work we present a robust ML potential for the EC:EMC binary solvent, a key component of liquid electrolytes in rechargeable Li-ion batteries. We identify the necessary ingredients needed to successfully model this liquid mixture of organic molecules. In particular, we address the challenge posed by the separation of scale between intra- and inter-molecular interactions, which is a general issue in all condensed phase molecular systems.

Empirically Optimized One-Electron Pseudopotential for the Hydrated Electron: A Proof-of-Concept Study.

Mixed quantum-classical molecular dynamics simulations have been important tools for studying the hydrated electron. They generally use a one-electron pseudopotential to describe the interactions of an electron with the water molecules. This approximation shows both the strength and weakness of the approach. On the one hand, it enables extensive statistical sampling and large system sizes that are not possible with more accurate ab initio molecular dynamics methods. On the other hand, there has (justifiably) been much debate about the ability of pseudopotentials to accurately and quantitatively describe the hydrated electron properties. These pseudopotentials have largely been derived by fitting them to ab initio calculations of an electron interacting with a single water molecule. In this paper, we present a proof-of-concept demonstration of an alternative approach in which the pseudopotential parameters are determined by optimizing them to reproduce key experimental properties. Specifically, we develop a new pseudopotential, using the existing TBOpt model as a starting point, which correctly describes the hydrated electron vertical detachment energy and radius of gyration. In addition to these properties, this empirically optimized model displays a significantly modified solvation structure, which improves, for example, the prediction of the partial molar volume.

Strategies for fitting accurate machine-learned inter-atomic potentials for solid electrolytes

Ion transport in materials is routinely probed through several experimental techniques, which introduce variability in reported ionic diffusivities and conductivities. The computational prediction of ionic diffusivities and conductivities helps in identifying good ionic conductors, and suitable solid electrolytes (SEs), thus establishing firm structure-property relationships. Machine-learned potentials are an attractive strategy to extend the capabilities of accurate ab initio molecular dynamics (AIMD) to longer simulations for larger systems, enabling the study of ion transport at lower temperatures. However, machine-learned potentials being in their infancy, critical assessments of their predicting capabilities are rare. Here, we identified the main factors controlling the quality of a machine-learning potential based on the moment tensor potential formulation, when applied to the properties of ion transport in ionic conductors, such as SEs. Our results underline the importance of high-quality and diverse training sets required to fit moment tensor potentials. We highlight the importance of considering intrinsic defects which may occur in SEs. We demonstrate the limitations posed by short-timescale and high-temperature AIMD simulations to predict the room-temperature properties of materials.

Solvent Effect on the Structure and Properties of RGD Peptide (1FUV) at Body Temperature (310 K) Using Ab Initio Molecular Dynamics.

The structure and properties of the arginine-glycine-aspartate (RGD) sequence of the 1FUV peptide at 0 K and body temperature (310 K) are systematically investigated in a dry and aqueous environment using more accurate ab initio molecular dynamics and density functional theory calculations. The fundamental properties, such as electronic structure, interatomic bonding, partial charge distribution, and dielectric response function at 0 and 310 K are analyzed, comparing them in dry and solvated models. These accurate microscopic parameters determined from highly reliable quantum mechanical calculations are useful to define the range and strength of complex molecular interactions occurring between the RGD peptide and the integrin receptor. The in-depth bonding picture analyzed using a novel quantum mechanical metric, the total bond order (TBO), quantifies the role played by hydrogen bonds in the internal cohesion of the simulated structures. The TBO at 310 K decreases in the dry model but increases in the solvated model. These differences are small but extremely important in the context of conditions prevalent in the human body and relevant for health issues. Our results provide a new level of understanding of the structure and properties of the 1FUV peptide and help in advancing the study of RGD containing other peptides.

Chemical Desorption versus Energy Dissipation: Insights from Ab Initio Molecular Dynamics of HCO· Formation

Abstract Molecular clouds are the cold regions of the Milky Way where stars form. They are enriched by rather complex molecules. Many of these molecules are believed to be synthesized on the icy surfaces of the interstellar submicron-sized dust grains that permeate the Galaxy. At 10 K thermal desorption is inefficient and, therefore, why these molecules are found in the cold gas has tantalized astronomers for years. The assumption of the current models, called chemical desorption, is that the molecule formation energy released by the chemical reactions at the grain surface is partially absorbed by the grain and the remaining energy causes the ejection of the newly formed molecules into the gas. Here we report accurate ab initio molecular dynamics simulations aimed at studying the fate of the energy released by the first reaction of the H· addition chain to CO, H· + CO HCO·, occurring on a crystalline ice surface model. We show that about 90% of the HCO· formation energy is injected toward the ice in the first picosecond, leaving HCO· with an energy content (10–15 kJ mol−1) of less than half its binding energy (30 kJ mol−1). As a result, in agreement with laboratory experiments, we conclude that chemical desorption is inefficient for this specific system, namely H· + CO on crystalline ice. We suspect this behavior to be quite general when dealing with hydrogen bonds, which are responsible for both the cohesive energy of the ice mantle and the interaction with adsorbates, as HCO·, even though ad hoc simulations are needed to draw specific conclusions on other systems.

The Quest for Accurate Liquid Water Properties from First Principles.

Developing accurate ab initio molecular dynamics (AIMD) models that capture both electronic reorganization and nuclear quantum effects associated with hydrogen bonding is key to quantitative understanding of bulk water and its anomalies as well as its role as a universal solvent. For condensed phase simulations, AIMD has typically relied on the generalized gradient approximation (GGA) of density functional theory (DFT) as the underlying model chemistry for the potential energy surface, with nuclear quantum effects (NQEs) sometimes modeled by performing classical molecular dynamics simulations at elevated temperatures. Here we show that the properties of liquid water obtained from the meta-GGA B97M-rV functional, when evaluated using accelerated path integral molecular dynamics simulations, display accuracy comparable to a computationally expensive dispersion-corrected hybrid functional, revPBE0-D3. We show that the meta-GGA DFT functionalreproduces bulk water properties including radial distribution functions, self-diffusion coefficients, and infrared spectra with comparable accuracy of a much more expensive hybrid functional. This work demonstrates that the underlying quality of a good DFT functional requires evaluation with quantum nuclei and that high-temperature simulations are a poor proxy for properly treating NQEs.

A Versatile Multiple Time Step Scheme for Efficient ab Initio Molecular Dynamics Simulations.

We present here our implementation of a time-reversible, multiple time step (MTS) method for full QM and hybrid QM/MM Born-Oppenheimer molecular dynamics simulations. The method relies on a fully flexible combination of electronic structure methods, from density functional theory to wave function-based quantum chemistry methods, to evaluate the nuclear forces in the reference and in the correction steps. The possibility of combining different electronic structure methods is based on the observation that exchange and correlation terms only contribute to low frequency modes of nuclear forces. We show how a pair of low/high level electronic structure methods that individually would lead to very different system properties can be efficiently combined in the reference and correction steps of this MTS scheme. The current MTS implementation makes it possible to perform highly accurate ab initio molecular dynamics simulations at reduced computational cost. Stable and accurate trajectories were obtained with time steps of several femtoseconds, similar to and even exceeding the ones usually adopted in classical molecular dynamics, in particular when using a generalized Langevin stochastic thermostat. Compared to the standard Velocity Verlet integration, the present MTS scheme allows for a 5- to 6-fold overall speedup, at an unaltered level of accuracy.

Assessing the accuracy of improved force-matched water models derived from Ab initio molecular dynamics simulations.

The accuracy of water models derived from ab initio molecular dynamics simulations by means on an improved force-matching scheme is assessed for various thermodynamic, transport, and structural properties. It is found that although the resulting force-matched water models are typically less accurate than fully empirical force fields in predicting thermodynamic properties, they are nevertheless much more accurate than generally appreciated in reproducing the structure of liquid water and in fact superseding most of the commonly used empirical water models. This development demonstrates the feasibility to routinely parametrize computationally efficient yet predictive potential energy functions based on accurate ab initio molecular dynamics simulations for a large variety of different systems. © 2016 Wiley Periodicals, Inc.

In search of zero thermal expansion anisotropy in Mo5Si3 by strategic alloying

Reducing the thermal expansion anisotropy (TEA) of alloy compounds is one of the most important issues for their potential applications in high temperature environment. The Mo5Si3 (T1 phase) is known to be an important intermetallic compound with high melting temperature. Unfortunately, its large TEA renders it unsuitable for high temperature structural/coating applications. Many attempts have been made in the past to reduce TEA by substituting Mo by other transition metal ions such as V with little success and some unexpected observations. Here we use accurate ab initio molecular dynamics (AIMD) simulations to obtain the TEA from thermal expansion coefficients for two T1 phase alloy systems (Mo,V)5Si3 and Mo5(Si,Al)3. We demonstrate that strategic alloying with Al substituting Si can achieve zero TEA for T1 phase. The microscopic origin of this outstanding thermomechanical properties in this alloy is explained by the calculation of higher order elastic constants in conjunction with atom and direction-resolved phonon density of states.

Accurate Static and Dynamic Properties of Liquid Electrolytes for Li-Ion Batteries from ab initio Molecular Dynamics

Lithium-ion batteries have the potential to revolutionize the transportation industry, as they did for wireless communication. A judicious choice of the liquid electrolytes used in these systems is required to achieve a good balance among high-energy storage, long cycle life and stability, and fast charging. Ethylene-carbonate (EC) and propylene-carbonate (PC) are popular electrolytes. However, to date, almost all molecular-dynamics simulations of these fluids rely on classical force fields, while a complete description of the functionality of Li-ion batteries will eventually require quantum mechanics. We perform accurate ab initio molecular-dynamics simulations of ethylene- and propylene-carbonate with LiPF(6) at experimental concentrations to build solvation models which explain available neutron scattering and nuclear magnetic resonance (NMR) results and to compute Li-ion solvation energies and diffusion constants. Our results suggest some similarities between the two liquids as well as some important differences. Simulations also provide useful insights into formation of solid-electrolyte interphases in the presence of electrodes in conventional Li-ion batteries.